I have some question about Alpamayo.
- When it will support stuff like routing/navigation, but also parking/unparking and navigating to the car owner based on visual owner description for example?
In the 1.5 version it mentions
"Alpamayo 1.5 is built on the [Cosmos-Reason2] VLM backbone, is RL post-trained, and introduces support for navigation guidance, flexible camera counts, and user question answering."
-
How does it support navigation guidance and user question answering please? What does that mean? Like I can navigate the model myself?
-
Will the minimum hardware requirements for such model be NVIDIA Jetson AGX Orin 64GB ?
-
Is the model able to realize that road is closed and it chooses alternative route?
-
Is the Alpamayo 1.5 already optimized to the model size, or it can be shrinked down even more to run on cheaper hardware like Nvidia Jetson Nano ?
-
How does the model handle frametime reaction speed, will it support the use of event cameras in the future or even LiDAR?
-
Is the model physics aware like a world model?
-
Does the model remember everything from the startup, or it knows when to forget temporal information and continues, or writes a logs of events, or it's like a conversation history and context window?
I have some question about Alpamayo.
In the 1.5 version it mentions
"Alpamayo 1.5 is built on the [Cosmos-Reason2] VLM backbone, is RL post-trained, and introduces support for navigation guidance, flexible camera counts, and user question answering."
How does it support navigation guidance and user question answering please? What does that mean? Like I can navigate the model myself?
Will the minimum hardware requirements for such model be NVIDIA Jetson AGX Orin 64GB ?
Is the model able to realize that road is closed and it chooses alternative route?
Is the Alpamayo 1.5 already optimized to the model size, or it can be shrinked down even more to run on cheaper hardware like Nvidia Jetson Nano ?
How does the model handle frametime reaction speed, will it support the use of event cameras in the future or even LiDAR?
Is the model physics aware like a world model?
Does the model remember everything from the startup, or it knows when to forget temporal information and continues, or writes a logs of events, or it's like a conversation history and context window?